Publications

Fast Spectrally Sparse Signal Reconstruction via Jacobi-Preconditioned Gradient Descent
Jian-Feng Cai
Xueyang Quan
Jiaxi Ying
Spectrally sparse signal reconstruction arises in a wide range of applications and can be formulated as a low-rank Hankel matrix completion … (see more)problem. We develop a Jacobi-preconditioned gradient descent method that preserves the low per-iteration complexity of first-order algorithms while achieving linear convergence at a rate independent of the condition number. By introducing a generator that maps factor-based iterates to matrix space, we establish equivalence with manifold-based methods, enabling direct convergence analysis while avoiding the need to define distances under complex-symmetric factorization ambiguity. Extensive experiments demonstrate that the proposed algorithm outperforms state-of-the-art methods in both iteration count and computational time across a broad range of problem settings.
Fewer Expert Transfers Without Changing Routing: Expert Eviction for MoE Offloading
Jean-Maxime Larouche
A mixture-of-experts (MoE) language model uses only a few experts per token, so a server can keep most expert weights in CPU memory and copy… (see more) an expert to the GPU when the router requests it. Recent methods train the router to keep the same experts for several tokens, and treat fewer routing switches as a proxy for fewer copies. We measure the trade-off directly, in a harness that performs and times real CPU-to-GPU copies for OLMoE-1B-7B. On an L40S GPU, every tested policy that holds an expert set fixed and decodes faster than the least-recently-used (LRU) cache increases next-token loss by at least 0.067 nats. A change that leaves every routing decision intact does better: when the cache must evict an expert, evict the one whose next use lies farthest in the future. This rule copies 2.1x fewer experts than LRU at no loss in quality and is within 3.5% of the fewest copies possible. It needs future requests; the next sixteen tokens of demand recover 96% of its reduction, whereas a rule based on counts of past requests recovers 27%.
GLaS-JEPA: Gaussian-Regularized Speech SSL without Engineered Prediction Targets
Samir Sadok
Thomas Hueber
Xavier Alameda-Pineda
Speech self-supervised learning aims to learn general-purpose representations for downstream speech tasks. However, current approaches rely … (see more)on complex, carefully designed prediction targets. We challenge this necessity with GLaS-JEPA, a framework that directly predicts the current encoder's continuous representations at masked positions, without contrastive learning, discrete targets, or separate EMA target encoders. We prevent representation collapse using SIGReg representation-space regularization, eliminating the need for engineered target-generation mechanisms. Pretrained on 960 hours of LibriSpeech, our 57M-parameter model achieves a 6.89% WER on frozen-encoder SUPERB ASR and a 25.87% CER on slot filling, outperforming the best non-distilled sub-90M baselines by 43.1% and 22.0%, respectively. These results demonstrate that highly competitive speech representations can emerge from a radically simplified training recipe.
Hierarchical Retrieval at Scale: Bridging Interpretability and Efficiency
Information retrieval is a core component of many intelligent systems as it enables conditioning of outputs on new and large-scale datasets.… (see more) While effective, the standard practice of encoding data into high-dimensional representations for similarity search entails large memory and compute footprints, and also makes it hard to inspect the inner workings of the system. Hierarchical retrieval methods offer an interpretable alternative by organizing data at multiple granular levels, yet do not match the efficiency and performance of flat retrieval approaches. In this paper, we propose Retreever, a tree-based method that makes hierarchical retrieval viable at scale by directly optimizing its structure for retrieval performance while naturally providing transparency through meaningful semantic groupings. Our method offers the flexibility to balance cost and utility by indexing data using representations from any tree level. We show that Retreever delivers strong coarse (intermediate levels) and fine representations (terminal level), while achieving the highest retrieval accuracy at the lowest latency among hierarchical methods. These results demonstrate that this family of techniques is viable in practical applications.
Improving LLM-Based Recommenders with Conservative Generative Flow Networks
Xuan Yu
Feng Niu
Rui Zhu
Yudong Zhang
Xu Wang
Generative Flow Networks (GFlowNets) have recently been used to improve diversity and mitigate popularity bias in LLM-based recommender syst… (see more)ems, yet most objectives are developed under online-style assumptions. In offline LLM-based recommendation, learning is constrained to a fixed logged dataset, yielding partial support over token transitions on the dataset-induced token-prefix DAG. Naively applying Sub-Trajectory Balance (SubTB) becomes non-identifiable and can arbitrarily allocate probability mass to unsupported regions. We formalize this failure and identify three sources of non-identifiability that induce distributional shift between the dataset-implied policy and the learned policy: (i) flow overestimation, (ii) forward mass leakage, and (iii) backward compensation. To address it, we propose CFlower, which introduces a conservative SubTB objective that explicitly penalizes unsupported forward flow mass, and combines it with dataset-constrained policy learning with on-policy sampling on the dataset-induced DAG for efficient training under offline constraints. Experiments on three Amazon recommendation datasets show that CFlower improves distributional matching and delivers a stronger accuracy–exposure trade-off than prior GFlowNet and SFT baselines, while serving as a more reliable reference policy for downstream RL fine-tuning.
Moving beyond 5-HT2A: language models map neurotransmitter receptors to experiential semantics
Josie Kins
Lex Levkovskyy
Andrew Thieme
Paul B. Wren
Tom Ray
Frederick Streeter Barrett
Robin Lester Carhart-Harris
Scaling Transformers for End-to-End Discrete Audio Tokenization
Yitian Gong
Kuangwei Chen
Zhaoye Fei
Xiaogui Yang
Ke Chen
Kexin Huang
Mingshu Chen
Ruixiao Li
Qinyuan Cheng
Shimin Li
Xipeng Qiu
Discrete audio tokenizers are fundamental to empowering large language models with native audio processing and generation capabilities. Desp… (see more)ite recent progress, existing approaches often rely on pretrained encoders, semantic distillation, or heterogeneous CNN-based architectures. These designs introduce fixed inductive biases that limit reconstruction fidelity and hinder effective scaling. In this paper, we argue that discrete audio tokenization should be learned fully end-to-end using a homogeneous and scalable architecture. Based on this perspective, we propose
Structural Remodeling of the Lumbar Spine Following Non-Surgical Spinal Decompression: An Automated MRI Study
Jacob A. Connolly
Kenneth A. Weber
Julien Cohen‐Adad
Nathan D. Schilaty
Background and Objectives: To investigate MRI changes using SpineReport (r20260928) following non-surgical spinal decompression (NSSD) in pa… (see more)tients with chronic low back pain (cLBP). We conducted a retrospective review of patients treated with NSSD who obtained MRI within 6 months of treatment initiation. Methods: Thirty-six subjects (20 F, 16 M), with a mean age of 55.9 ± 16.4 years, were treated at the USF Health Chiropractic Clinic, Department of Neurosurgery, Brain & Spine. NSSD using DRX9000® (Excite Medical, Tampa, FL, USA) was utilized, with a median of 20 (20, 25) treatments. Treatment levels included L3 (n = 2), L4 (n = 18), and L5 (n = 16). Sagittal T2-weighted images were processed using TotalSpineSeg (r20251124) for automated segmentation, with SpineReport used to quantify disc, vertebrae, spinal canal, and intervertebral foramen metrics. Results: Mean pain scores improved by 3.4 ± 1.8 from 5.6 ± 1.6 (p 0.001), while Oswestry Disability Index scores improved by 10.8 ± 9.1 from 17.3 ± 8.0 (p 0.001). At the treated level, spinal canal cross-sectional area increased from 170.9 to 175.8 mm2 (2.9%, p = 0.040), and right–left diameter increased from 16.0 to 16.3 mm (1.9%, p = 0.025). Disc volume decreased from 7762.8 to 7449.5 mm3 (−4.0%, p = 0.046). No significant changes were observed in disc thickness, DHI, intervertebral foramen area, or vertebral measurements (p > 0.05). Conclusions: SpineReport enabled automated quantification of key spinal structures. NSSD was associated with increased spinal canal dimensions and improved symptoms. Further research is needed to determine the persistence and clinical significance of these structural changes.
SymSpectra: Symmetric Information Bottleneck Framework for Molecular Structure Recognition under Imbalanced Settings
Xiaohan Qin
Wenjie Du
Identifying molecular structures from spectral data is essential for early-stage chemical analysis, yet it remains a difficult task due to s… (see more)evere functional group imbalance and complex inter-group dependencies, which often cause existing methods to overfit frequent groups while underperforming on rare ones. To address these issues, we present SymSpectra, a Symmetric Conditional Information Bottleneck (SCIB) framework designed to seamlessly integrate multi-modal Spectra features. Our model employs the SCIB framework to fuse multi-modal spectroscopic data into a unified representation, effectively preserving discriminative signals while mitigating redundancy. To enhance robustness against data imbalance, we incorporate conditional mutual information into the training objective, increasing the model’s sensitivity to rare functional groups and challenging molecular cases. Additionally, a specialized module captures the dependencies among functional groups, improving both prediction accuracy and chemically meaningful interpretability. Experiments on multimodal spectral datasets show that SymSpectra outperforms state-of-the-art methods, achieving an F1-score of 0.970 in substructure classification and demonstrating strong robustness under various imbalance settings.
Topology-Aware Contrastive Learning: Regulating Representation Connectivity via Persistent Homology
Jiaxin Sun
Yuhua Qian
Standard contrastive learning minimizes geometric distance between positive pairs, implicitly assuming that strict compactness optimizes dis… (see more)crimination. However, this topology-agnostic confusion neglects intrinsic data structures and topological complexity, leading to class confusion—particularly when aggressive augmentations induce semantic drift. To address this, we propose Topology-Aware Contrastive Learning, a framework that shifts the objective from geometric singularity to topological connectivity. Leveraging Persistent Homology, we explicitly regulate the connectivity of the latent space, ensuring positive pairs maintain an
BiomiX 3.0: A user-friendly platform for democratized multi-omics integration with graph-based learning
Iñigo Clemente‐Larramendi
Jessica Gliozzo
Lourdes Velo‐Suárez
Ozvan Bocher
Nathan Foulquier
Maria Hernandez-Valladares
Marc‐André Legault
Divi Cornec
Christophe Jamin
Alberto Gil-de-la-Fuente
Álvaro Fernández‐Ochoa
Cristian Iperi
Abstract Background Multi-omics integration has emerged as a powerful strategy to decode the molecular complexity of biological systems. How… (see more)ever, the diversity of available methods, each designed with distinct assumptions, objectives, and computational requirements, makes method selection, usage and interpretation challenging for non-expert users. Here we present BiomiX 3.0, an updated version of the BiomiX platform that extends its integration capabilities with four additional methods: Similarity Network Fusion (SNF), NEighborhood-based Multi-Omics clustering (NEMO), Data Integration Analysis for Biomarker discovery using Latent variable approaches for Omics studies (DIABLO), and PRAMIGO (Phenotyping netwoRk Application for Multi-omics InteGratiOn), a novel supervised heterogeneous graph transformer (HGT) introduced in this work. Results We benchmarked all five methods, MOFA, DIABLO, SNF, NEMO, and PRAMIGO, on two independent multi-omics datasets derived from a Chronic Lymphocytic Leukemia (CLL) cohort comparing IGHV-mutated and unmutated patients, and a pulmonary tuberculosis (PTB) cohort versus healthy controls. Supervised methods (DIABLO, PRAMIGO) consistently achieved higher condition-specific discrimination as measured by the Adjusted Rand Index (ARI) and the Adjusted Mutual Information (AMI). In contrast, unsupervised methods (SNF, NEMO) revealed alternative patient stratifications driven by independent sources of biological variance while MOFA performed in a semi-supervised way occupies an intermediate position, capturing latent factors that explain both disease-associated and orthogonal sources of variance. Systematic gene-centric analysis of the top-ranked features prioritized by each method was supported by manual biological annotation of shared and method-specific signals. Across cohorts, we annotated 154 shared features (116 genes, 38 metabolites) and 60 method-unique features per cohort, demonstrating that no single integration strategy captures the full landscape of biologically relevant signals. In the CLL cohort, shared features spanned B-cell receptor biology, innate immune signaling, RAS/MAPK activation, and epigenetic regulation, while method-unique features revealed supervised-method-specific insights into vesicle trafficking (DIABLO), immune checkpoints (MOFA), and ncRNA regulation (PRAMIGO). In the PTB cohort, a convergent interferon/innate immune signature dominated shared features across all methods. Still, method-unique analysis uncovered DIABLO-specific acylcarnitine metabolic reprogramming, MOFA-specific restoration of lysosomal trafficking, and PRAMIGO-specific γδ T-cell and immunoglobulin repertoire diversity. Across both cohorts, SNF and NEMO proved useful for detecting biological and technical sources of variation that were orthogonal to the primary condition of interest. Ultimately, PRAMIGO uniquely enables the construction of heterogeneous graphs modeling cross-modal molecular interactions, uncovering epigenetic co-regulation programs in CLL and multi-omics inflammatory modules in PTB that are difficult to identify using conventional integration approaches. Conclusions BiomiX 3.0 provides a graphical user interface (GUI) multi-method integration environment that democratizes access to state-of-the-art multi-omics analysis. By combining both unsupervised and supervised integration strategies within a unified platform and introducing graph-based learning through PRAMIGO, BiomiX 3.0 enables researchers across disciplines with complementary tools to interrogate the biological sources of variation in their data, without requiring bioinformatics expertise.
Fair Policy Optimization in Major-Minor Weakly Coupled Markov Decision Processes
We consider fair resource allocation in sequential decision-making environments modeled as major-minor weakly coupled Markov decision proces… (see more)ses (M2WCMDP). In this framework, resource constraints couple the action spaces of a major sub-Markov decision process (sub-MDP) and a population of minor sub-MDPs that would otherwise operate independently. Instead of using the traditional utilitarian (total-sum) objective, we optimize a general class of monotone, concave, permutation-invariant, normalized fairness functions. With homogeneous minor sub-MDPs, we prove that the problem under symmetry reduces to optimizing the platform-plus-mean-participant utilitarian objective over the class of \textit{permutation-invariant} policies, which allows us to exploit efficient algorithms that optimize the utilitarian-based objective to solve this fairness-aware problem. For more general settings, we introduce a count-proportion-based deep reinforcement learning approach with a priority-based sampler that generates feasible count actions. The generality of our framework means that the proposed algorithms and theoretical guarantees transfer to any domain with a symmetric M2WCMDP structure. We consider two applications: the machine replacement problem and the joint control of pricing and taxi relocation problem on a New York City-calibrated dataset. We validate our theoretical findings with comprehensive experiments, confirming the effectiveness of our proposed method in achieving strong fairness-aware performance while remaining scalable.